KEY TAKEAWAYS
- The study aimed to assess the impact of lung nodule detection by AI and HR.
- AI produced more false positives per scan in people with emphysema compared to those without, a difference not seen with HR.
Emphysema, a lung disease that damages lung tissue, can be identified on CT scans. These scans are also used to detect lung cancer, and the presence of emphysema may influence how lung tissue appears in both contexts.
Nikos Sourlos and the team aimed to evaluate if this impacts lung nodule detection by artificial intelligence (AI) and human readers (HR).
Participants were drawn from the “Lifelines” cohort and had received low-dose chest CT scans. Nodules in individuals without emphysema were matched with similar-sized nodules in those with at least moderate emphysema.
AI findings for nodules measuring 30-100 mm3 and 101-300 mm3 were compared with HR results. Two expert radiologists independently reviewed any discrepancies. Sensitivity and false positives (FPs) per scan were compared between the emphysema and non-emphysema groups.
Among the 121 participants, 39 individuals with emphysema and 82 without, aged 61 ± 8 years, and 58 males (47.9%), both AI and HR detected nodular findings. Specifically, AI detected 196 nodular findings, while HR identified 206, resulting in 109 concordant nodules and 184 discrepancies, including 118 true nodules.
AI demonstrated a sensitivity of 0.68 (95% CI 0.57-0.77) in individuals with emphysema, compared to 0.71 (0.62-0.78) in those without emphysema. The false positives per scan (FPs/scan) were 0.51 and 0.22, respectively (P= 0.028). For HR, the sensitivity was 0.76 (0.65-0.84) and 0.80 (0.72-0.86), with FPs/scan of 0.15 and 0.27, respectively (P= 0.230).
Overall, HR exhibited slightly higher sensitivity compared to AI. However, this difference vanished after excluding benign lymph nodes. Notably, FPs/scans were significantly higher for AI in emphysema compared to non-emphysema (P= 0.028). Conversely, FPs/scans for HR were higher than AI for nodules measuring 30-100 mm3 in non-emphysema cases (P= 0.009).
The study concluded that AI exhibited a higher rate of false positives per scan in individuals with emphysema compared to those without, a distinction not evident for HR.
Funding was provided by the Netherlands Organisation for Scientific Research/Innovative Medical Device Initiative, Siemens Healthineers, the Dutch Ministry of Health, Welfare, and Sport, the Dutch Ministry of Economic Affairs, the University Medical Center Groningen (UMCG), Groningen University, and the Provinces in the North of the Netherlands.
Source: https://pubmed.ncbi.nlm.nih.gov/38764066/
Sourlos N, Pelgrim G, Wisselink HJ, et al. (2024). “Effect of emphysema on AI software and human reader performance in lung nodule detection from low-dose chest CT.” Eur Radiol Exp. 2024 May 20;8(1):63. doi: 10.1186/s41747-024-00459-9. PMID: 38764066.
